Papers

New on arXiv

Fri 1 May 2026

371 papers

Just Keep Prompting: Evaluating Repetitive Socratic Prompting in VLMsLLM reasoning → Multimodal reasoning +1
Breaking Safety at the Token Boundary: How BPE Tokenization Creates Exploitable Gaps in LLM AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
KARA: Efficient Reasoning LLM Serving via Sliding-Window KV Cache CompressionLLM reasoning → Test-time compute and efficient thinking +1
Safeguarding LLM Agents from Misalignment through Provenance AnalysisLLM agents and tool use → Agent threats, guardrails and governance
Beyond Compaction: Structured Context Eviction for Long-Horizon AgentsLLM agents and tool use → Agent memory systems
Human-AI Coordination Zones: A Framework for Designing Human-in-the-Loop Experiences with Agentic AILLM agents and tool use → General LLM agent frameworks and benchmarks
Multimodal Large Language Models as Synthetic Participants in Video-Based Studies: An EvaluationMultimodal LLMs / vision-language models → Multimodal emotion understanding
TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM CoordinationLLM agents and tool use → General LLM agent frameworks and benchmarks
AgentStop: Terminating Local AI Agents Early to Save Energy in Consumer DevicesLLM agents and tool use → Software engineering coding agents
AgenticAITA: A Proof-Of-Concept About Deliberative Multi-Agent Reasoning for Autonomous Trading SystemsLLM reasoning → Reasoning for agents, planning and code +1
DARE: Diffusion Language Model Activation Reuse for Efficient InferenceDiffusion models → Discrete and text diffusion
VLADriver-RAG: Retrieval-Augmented Vision-Language-Action Models for Autonomous DrivingVision-language-action models → VLA for autonomous driving +2
Interactive Inverse Reinforcement Learning of Interaction Scenarios via Bi-level OptimizationOffline RL and imitation learning from data → Inverse reinforcement learning +1
Topology-Driven Anti-Entanglement Control for Soft RobotsDeep reinforcement learning methods → RL for robot locomotion and manipulation
RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
Healthcare AI GYM for Medical AgentsKnowledge distillation → On-policy and self-distillation for LLMs
PrismAgent: Illuminating Harm in Memes via a Zero-Shot Interpretable Multi-Agent FrameworkLLM safety and security → Safety alignment, refusal and guardrails
Proteo-R1: Reasoning Foundation Models for De Novo Protein DesignLLM reasoning → Clinical and biomedical applications
Minimizing Collateral Damage in Activation SteeringMechanistic interpretability → Activation steering methods
Position: Safety and Fairness in Agentic AI Depend on Interaction Topology, Not on Model Scale or AlignmentLLM agents and tool use → Agent threats, guardrails and governance
A Low-Latency Fraud Detection Layer for Detecting Adversarial Interaction Patterns in LLM-Powered AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
ScribbleEdit: Synthetic Data for Image Editing with Scribbles and TextSynthetic data and self-improvement for LLMs → Self-improving MLLMs and synthetic data
When Embedding-Based Defenses Fail: Rethinking Safety in LLM-Based Multi-Agent SystemsLLM reasoning → Reasoning for agents, planning and code +1
Forager: a lightweight testbed for continual learning with partial observability in RLDeep reinforcement learning methods → Deep RL algorithms and representations
Iterative Finetuning is Mostly IdempotentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Revisiting Privacy Leakage in Machine Unlearning: Membership Inference Beyond the Forgotten SetLLM safety and security → Machine unlearning
PERSA: Reinforcement Learning for Professor-Style Personalized Feedback with LLMsLLM alignment / RLHF / preference optimisation → Process reward models for reasoning
Disciplined Diffusion: Text-to-Image Diffusion Model against NSFW GenerationDiffusion models → Robustness, privacy and adversarial aspects
Component-Aware Self-Speculative Decoding in Hybrid Language ModelsEfficient LLM inference and serving → Speculative decoding
Towards Multi-Agent Autonomous Reasoning in HydrodynamicsLLM reasoning → Reasoning for agents, planning and code +1
Virtual Speech Therapist: A Clinician-in-the-Loop AI Speech Therapy Agent for Personalized and Supervised TherapyLLM agents and tool use → Medical and scientific domain agents
Learning to Race in Minutes: Infoprop Dyna on the Mini WheelbotDeep reinforcement learning methods → RL for autonomous driving and traffic
A Sentence Relation-Based Approach to Sanitizing Malicious InstructionsRetrieval-augmented generation → Privacy and security of RAG +1
PRISM: Exposing and Resolving Spurious Isolation in Federated Multimodal Continual LearningMixture-of-experts → Mixture of LoRA experts
LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer InferenceEfficient LLM inference and serving → Early exit and edge inference
LLM Ghostbusters: Surgical Package Hallucination Suppression via Adaptive UnlearningLLM safety and security → Backdoor and data poisoning attacks
A Theoretical Game of Attacks via Compositional SkillsLLM safety and security → Prompt injection and agent security
SRTJ: Self-Evolving Rule-Driven Training-Free LLM JailbreakingLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
MedMosaic: A Challenging Large Scale Benchmark of Diverse Medical AudioLLM reasoning → Clinical and biomedical applications
"I Don't Know" -- Towards Appropriate Trust with Certainty-Aware Retrieval Augmented GenerationRetrieval-augmented generation → RAG robustness and adaptive retrieval
E-MIA: Exam-Style Black-Box Membership Inference Attacks against RAG SystemsRetrieval-augmented generation → Privacy and security of RAG +1
ARIS: Agentic and Relationship Intelligence System for Social RobotsWorld models → World models for language agents +1
PPO guided Agentic Pipeline for Adaptive Prompt Selection and Test Case GenerationLLM agents and tool use → Software engineering coding agents
Uncertainty in a Single Pass: A Closed-Form Identity for One-Step Flow MatchingDiffusion models → Flow matching, time series and science
Fusing Urban Structure and Semantics: A Conditional Diffusion Model for Cross-City OD Matrix GenerationDiffusion models → Diffusion and flow policies for robotics
Watch Your Step: Information Injection in Diffusion Models via Shadow Timestep EmbeddingDiffusion models → Diffusion foundations and sampling
Code World Model Preparedness ReportWorld models → World models for language agents
Posterior Augmented Flow MatchingDiffusion models → Flow matching, time series and science
Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Let ViT Speak: Generative Language-Image Pre-trainingMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Can Coding Agents Reproduce Findings in Computational Materials Science?LLM agents and tool use → Medical and scientific domain agents
GMGaze: MoE-Based Context-Aware Gaze Estimation with CLIP and Multiscale TransformerMixture-of-experts → Vision mixture-of-experts
When RAG Chatbots Expose Their Backend: An Anonymized Case Study of Privacy and Security Risks in Patient-Facing Medical AIRetrieval-augmented generation → Privacy and security of RAG
Make Your LVLM KV Cache More LightweightMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
Repurposing Image Diffusion Models for Adversarial Synthetic Structured Data: A Case Study of Ground Truth DriftDiffusion models → Image generation, text-to-image and editing
SAVGO: Learning State-Action Value Geometry with Cosine Similarity for Continuous ControlDeep reinforcement learning methods → RL theory and sample complexity
Themis: Training Robust Multilingual Code Reward Models for Flexible Multi-Criteria ScoringLLM reasoning → Reasoning for agents, planning and code +1
Position: agentic AI orchestration should be Bayes-consistentLLM agents and tool use → General LLM agent frameworks and benchmarks
Self-Adaptive Multi-Agent LLM-Based Security Pattern Selection for IoT SystemsLLM agents and tool use → Agent threats, guardrails and governance
To Call or Not to Call: A Framework to Assess and Optimize LLM Tool CallingLLM agents and tool use → Tool learning and function calling
EASE: Federated Multimodal Unlearning via Entanglement-Aware Anchor ClosureLLM safety and security → Machine unlearning
PhysEdit: Physically-Consistent Region-Aware Image Editing via Adaptive Spatio-Temporal ReasoningLLM reasoning → Multimodal reasoning
FinSafetyBench: Evaluating LLM Safety in Real-World Financial ScenariosLLM safety and security → Safety alignment, refusal and guardrails
STARE: Step-wise Temporal Alignment and Red-teaming Engine for Multi-modal Toxicity AttackLLM safety and security → Vision-language and text-to-image safety
BabelSafe: A Policy-Grounded Multilingual Safety Benchmark for LLMsLLM safety and security → Safety alignment, refusal and guardrails
Eliminating Hidden Serialization in Multi-Node Megakernel CommunicationMixture-of-experts → Efficient MoE training and inference systems
Evaluating the Architectural Reasoning Capabilities of LLM Provers via the Obfuscated Natural Number GameLLM reasoning → Mathematical reasoning
Augmented Lagrangian Multiplier Network for State-wise Safety in Reinforcement LearningDeep reinforcement learning methods → Safe and constrained RL
InpaintSLat: Inpainting Structured 3D Latents via Initial Noise OptimizationDiffusion models → 3D, point cloud and scene generation
UniVidX: A Unified Multimodal Framework for Versatile Video Generation via Diffusion PriorsDiffusion models → Video generation
Reinforcement Learning with Markov Risk Measures and Multipattern Risk ApproximationDeep reinforcement learning methods → RL theory and sample complexity
Learn where to Click from Yourself: On-Policy Self-Distillation for GUI GroundingLLM agents and tool use → GUI and mobile agents +1
Unlearning Offline Stochastic Multi-Armed BanditsLLM safety and security → Machine unlearning
BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code SynthesisRetrieval-augmented generation → RAG for code
H-RAG at SemEval-2026 Task 8: Hierarchical Parent-Child Retrieval for Multi-Turn RAG ConversationsRetrieval-augmented generation → Conversational and speech RAG
Recovering Hidden Reward in Diffusion-Based PoliciesOffline RL and imitation learning from data → Inverse reinforcement learning +1
LLM-Emu: Native Runtime Emulation of LLM Inference via Profile-Driven SamplingEfficient LLM inference and serving → Serving systems and request scheduling
Affinity Is Not Enough: Recovering the Free Energy Principle in Mixture-of-ExpertsMixture-of-experts → Expert routing strategies
Intrinsic Gradient Suppression for Label-Noise Prompt Tuning in Vision-Language ModelsMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Jailbreaking Vision-Language Models Through the Visual ModalityMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +2
Stable-GFlowNet: Toward Diverse and Robust LLM Red-Teaming via Contrastive Trajectory BalanceLLM safety and security → Prompt injection and agent security
A11y-Compressor: A Framework for Enhancing the Efficiency of GUI Agent Observations through Visual Context Reconstruction and Redundancy ReductionLLM agents and tool use → GUI and mobile agents
Colorful-Noise: Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image GenerationDiffusion models → Image generation, text-to-image and editing
Hierarchical Abstract Tree for Cross-Document Retrieval-Augmented GenerationRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU ClustersLLM agents and tool use → Software engineering coding agents
SpaceMoE: Realizing Distributed Mixture-of-Experts Inference over Space NetworksMixture-of-experts → Efficient MoE training and inference systems
Scale-Aware Adversarial Analysis: A Diagnostic for Generative AI in Multiscale Complex SystemsDiffusion models → Flow matching, time series and science
LLM-Oriented Information Retrieval: A Denoising-First PerspectiveRetrieval-augmented generation → RAG robustness and adaptive retrieval +1
Leveraging Vision-Language Models as Weak Annotators in Active LearningMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
From Local to Global to Mechanistic: An iERF-Centered Unified Framework for Interpreting Vision ModelsMechanistic interpretability → Interpreting vision and multimodal models
CleanBase: Detecting Malicious Documents in RAG Knowledge DatabasesRetrieval-augmented generation → Privacy and security of RAG +1
Soft Graph Diffusion Transformer for MIMO DetectionDiffusion models → Diffusion for compression and wireless
Scaling Video Understanding via Compact Latent Multi-Agent CollaborationMultimodal LLMs / vision-language models → Video understanding with MLLMs +1
Thinking in Text and Images: Interleaved Vision--Language Reasoning Traces for Long-Horizon Robot ManipulationLLM reasoning → Spatial and embodied reasoning
LIMSSR: LLM-Driven Sequence-to-Score Reasoning under Training-Time Incomplete Multimodal ObservationsLLM reasoning → Multimodal reasoning
AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement LearningLLM agents and tool use → Reinforcement learning for LLM agents
Skills as Verifiable Artifacts: A Trust Schema and a Biconditional Correctness Criterion for Human-in-the-Loop Agent RuntimesLLM agents and tool use → Agent threats, guardrails and governance
GD4: Graph-based Discrete Denoising Diffusion for MIMO DetectionDiffusion models → Diffusion for compression and wireless
Rethinking LLM Ensembling from the Perspective of Mixture ModelsLLM reasoning → Test-time compute and efficient thinking
Learning While Deploying: Fleet-Scale Reinforcement Learning for Generalist Robot PoliciesVision-language-action models → Reinforcement learning and post-training +1

All 371